Recognition Based On Fourier Descriptor For Images From Television Guidance

The normalized Fourier descriptors as global feature, aspect ratio and invagination as partial feature are abstracted with the method of discretizing image boundary in the research of plane images from TV guidance.First, actual grey images are filtered and segmented with threshold value.The main plane figure can be drawn from the received images which are processed in morthology.The figure edge can be detected and the figure boundary can be searched, tracked. During the procedure, a group of boundary coordinates are received. Hu geometrical invariants normalized Fourier descriptors are calculated. According to experiments of same type an different types,the normalized Fourier descriptors by discreted with fixed quantity 256 are chosen as global feature by comparing discrete rate.Aspect ratio and invagination value are chosen as partial features by comparing usual methods. Aiming at two usual plane models, principal axis is searched in segmentated images where method of furthest coordinate and symmetry axis theory are combined. A three dimensional vector constructed by aspect ratio and invagination value can be received. Combinating FD nine dimentional LT heft and three dimentional vectors, a combinational twelve dimentional feature vector is constructed.At last, combinational feature vector examples are classified and tested with LVQ net which is trained. The experiments explained defective plane models can be recognized better with the method combinating Fourier descriptors and partial features than the method of whole Fourier descriptor..